1 Executive Summary
1.1 Summary of Changes in Assessment Inputs
List of changes (if any) in the input data, including estimated catches assumed for the current year and projected catches for current year + 1 and current year + 2. List of changes (if any) in the assessment methodology. This is one of the most important sections of the SAFE report. Common mistakes in this section include: 1) listing something that has not changed, and 2) not listing something that has changed.
1.2 Summary of Results
This presentation lacks any new data and mainly focusses on issues raised by the SSC and includes some model development changes. In particular, we enhanced the capability of the full-feedback loop by modifying the code to work as an operating model. We then linked this to a test comparing a management procedure that has some merit given a critique on the nature of the stock-recruitment relationship (SRR) conforming to the available data. We note that some new research has indicated that the value of \(\sigma_R\) can be reasonably well estimated in bot traditional stock assessment models and in state-space versions. However, we caution against strict adherence to the estimates. This is because in the case for pollock, the SRR is heavily influenced by data to the right of \(B_{MSY}\) instead of nearer the origin (where measures of “steepness” might most reliably be estimated). The SRR is thus linked to the Tier 1 fishing mortality recommendation in two ways: 1) by the assumption about \(\sigma_R\) (smaller values increase the “precision” of \(F_{MSY}\)) and 2) by having most of the data far from the origin and hence having steepness affected observations that are distant from the origin.
1.3 Responses to SSC and Plan Team Comments
1.3.1 From the 2023 SSC minutes:
The SSC would prefer not to make a risk table adjustment based on the difference from Tier 1 to Tier 3 again during the 2024 assessment cycle. The SSC requests that the next stock assessment bring back a new approach that may include development of a constant buffer based on factors extrinsic to the stock assessment (ecosystem function), or a better representation of the uncertainty in the Tier 1 and control rule calculations such that a reduction from maximum ABC is not needed every year.
- Will approach this with a discussion of alternative HCRs for pollock that consider ecosystem function. Re-examine what was done in 2011, consider a proper alternative feedback approach, noting that the early work including climate change.
Use posterior distributions from the MCMC to determine probabilities in the risk table and expand the columns in the risk table to include the recommended ABC (and potentially higher values).
- Just requires some code modifications to do the computations off of the posteriors
Identify where MLE estimates are being used and where MCMC estimates are being used. Also see the SSC’s General Stock Assessment Comments to include convergence diagnostics any time Bayesian results are reported. If MCMC diagnostics continue to appear adequate, reference points could be calculated using the posterior distribution used, rather than an analytical calculation.
- MLE is used and asymptotic approximations are used everywhere for advice, MCMC has been used as a comparison only and to show inter-relationships. Just requires some code modifications to do the computations off of the posteriors
The SSC recommends that consideration be given to removal of the Japanese fishery CPUE index (1965-76) from the assessment, because this data set no longer seems to contribute to the assessment. A sensitivity test should be done to evaluate the effects of data removal on the assessment.
- Will investigate. In 1964 data suggests all young small fish from the north, these data are handed down, issues about catching being not represenative of population, doesn’t impact much, makes it hard to get different assessment models running so may make sense to remove
Catch-at-age data provided by foreign fishing agencies in the pre-Magnuson era were not produced using the same aging criteria as the AFSC age-and-growth program. Consideration should be given to removal of these data from the assessment. A sensitivity test should be done to evaluate the effects of data removal on the assessment.
- Will investigate, age-determination methods from historical records and/or Russian scientists.
Document the method used for determining the selectivity to use in the forward projections and continue to evaluate projection variability due to selectivity. The SSC appreciates the selectivity retrospective comparison and suggests that it might be helpful to limit the comparison to the projection used in each year against only the most recent (best) estimate of selectivity for that year.
- Better document selectivity assumptions for projections. Look at Cole’s approach for plotting out performance. Consider historical uncertainty and variability.
The SSC supports the use of posterior predictive distributions, an underutilized tool in fisheries science, but common in other fields. To fully implement this approach to Bayesian model checking the SSC recommends plotting a histogram for each data source of the percentile of the predictive distribution in which each data point lies, noting that in a highly consistent model this histogram would be uniform.
- This conclusion might be incorrect. Likely to write that this will evolve into better approaches to include in the future
There is an apparent shift towards older ages in fisheries and trawl survey selectivity that should be investigated further.
- Will investigate, focused around changes in availabilty.
The SSC agrees with the BSAI GPT’s proposal in their presentation to move the multi-species model out of the pollock stock assessment, where it has been included as an appendix since it was first developed. Instead, they suggested it would be a separate chapter listed in parallel with the ESR, as it applies to multiple stocks and informs the ESRs.
- Technically it’s presented as part of the BSAI assessment. Agree that it should be highlighted on its own.
The SSC suggests revisiting the treatment of the stock-recruit relationship in the assessment model using recent improvements in modeling approaches and a longer time series that encompasses the recent warm period in the EBS. Recruitment deviates should be from the stock-recruit relationship and should model variability among annual recruitment estimates based on information in the data and residual variability. The estimation process should ensure that log-normally distributed recruitments are mean unbiased, resulting in unbiased biomass estimates. If an informative prior is used for steepness, it should be based on a meta-analysis of related species and reflect the uncertainty of that meta-analysis. Further consideration of time periods (as in previous analyses) and the influence of temperature on the stock-recruit relationship may be helpful. The SSC recognizes that there were significant recent analyses in 2016, 2018 and 2020 and is not requesting a repeat of those but a review of previous work would be helpful.
- We examine a model where the SRR applies over all age classes along with a reevaluation of a number of factors that affect the SRR and the \(F_{MSY}\) values. We include a few model results where temperature is in the condition of the SRR. We also note that an evaluation of different ‘regimes’ or periods of low or higher recruitment is presented as part of the standard assessment presention.
2 Stock-recruit relationship sensitivities
An SSC comment was:
- The SSC suggests revisiting the treatment of the stock-recruit relationship in the assessment model using recent improvements in modeling approaches and a longer time series that encompasses the recent warm period in the EBS. Recruitment deviates should be from the stock-recruit relationship and should model variability among annual recruitment estimates based on information in the data and residual variability. The estimation process should ensure that log-normally distributed recruitments are mean unbiased, resulting in unbiased biomass estimates. If an informative prior is used for steepness, it should be based on a meta-analysis of related species and reflect the uncertainty of that meta-analysis. Further consideration of time periods (as in previous analyses) and the influence of temperature on the stock-recruit relationship may be helpful. The SSC recognizes that there were significant recent analyses in 2016, 2018 and 2020 and is not requesting a repeat of those but a review of previous work would be helpful.
We attempt to address this in the subsequent sections. For backgound, the FMP specifies that the SSC’s criterion for Tiers 1 and 2 ABC/OFL estimates depend on having reliable estimates of \(F_{MSY}\). For this reason, the stock-recruitment relationship (SRR) is a key component of the advice. Over the years, we have compared different aspects of the SRR assumptions. The SSC requested a further evaluation and recap of what has been done previously. The following aspects of the SRR were evaluated and reviewed this year
- Selectivity
- Time series length
- Temperature
- Priors (on steepness)
- Form (e.g., Ricker versus Beverton-Holt)
- \(\sigma_R\)
2.1 Evaluating the impact of selectivity assumptions
on stock recruitment relationships (SRR)
To examine the assumptions about fishery selectivity variability we ran alternative model configurations where selectivity variability was contrasted. In one configuration it was constrained such that the fishing mortality was considered completely separable with respect to age and one where there was limited constraint on the selectivity. This later model is similar in nature to traditional VPA models where the catch at age is assumed known precisely. The resulting selectivity patterns are shown in Figure 1. Results showed that very little difference between a freely specified selectivity model (Figure 2) Results comparing the constrained selectivity differed substantially from last year’s configuration (Figure 3).
The large differences due to assuming separability also impacts the estimates of the stock recruitment relationship (Figure 4). This figure also dipicts the different magnitude of the recent recruitment and an increased uncertainty. In particular, the 2018 year class is estimated to be much larger in the separable model.
Another form of evaluating the selectivity estimates was to simply apply each of the annual selectivity estimates (or partial Fs) from the past 20 years. We note that the base result used the mean selectivity over the recent 2 years; specifically, the mean selectivity for 2021 and 2022 for the 2023 terminal year assessment.
Results show that …xxx
2.2 Time series length for the stock-recruitment relationship conditioning
The SSC requested that the full time series be included for the SRR conditioning. Extending the time series back to 1964 resulted in a different shaped curve with higher steepness (Figure 5). This was due to the inclusion of age-1 recruits from the early years. The inclusion of sea-surface temperature as a covariate to the full time series moderated this increase slightly (Figure 7).
As another consideration for Tier 1 and Tier 3, we recognized that the SPR rate for the 2023 assessment that corresponded to a value of about \(F_{32\%}\). For contrast we used a similar SRR curve condition approach asking the question “what SRR satisfies the constraint that out \(F_{35\%}\) = \(F_{MSY}\).”
An alternative SRR conditioning exercise was conducted where the year range for the conditioning of the curve was dropped in successive years. This was intended to show how sensitive the curve is to the years included in the analysis. We expect that it should revert to the prior as fewer years are included in the conditioning of the SRR curve.
Non-stationarity XXX xxx
2.3 Including temperature effects in the stock-recruitment relationship
2.4 Removing the impact of the prior on the stock-recruitment relationship
2.5 Beverton-Holt versus Ricker stock-recruitment relationship
2.6 Specified variability about the SRR
The SSC requested that we ensure that the bias correction is applied in the application of fitting the SRR. We confirm that in past assessments, for the period which the SRR curve was applied, the bias correction term was included. Specifically,
\[ \hat{R}_t = f(B_{t-1}) e^{\epsilon_t-0.5 \sigma_R^2} \]
where \(\epsilon_t \sim N(0, \sigma_R^2)\). Here the production function (which generates age-1 recruitment) is a function of spawning biomass in the previous year (\(f(B_{t-1})\)). Since this function “generates” age-1 recruitment from a lognormal distribution, the basis for this must be scaled accordingly. Therefore the assessment model numbers-at-age one (\(\dot{R_t}=N_{1,t}\)), must account for the bias based on the SRR. This leads to the recruitment component of the negative log-likelihood as
\[ -ln(L_{rec}) = \sum_{t=1}^{T} \left[\frac{\left(\chi_t + \frac{\sigma^2_R}{2}\right)^2}{2\sigma^2_R} + ln(\sigma_R) \right] \]
where \(\chi_t = \log(\dot{R}_t) - \log(\hat{R}_t)\). Note that the bias correction term falls within the likelihood because the bias is a function of the model estimates.
For this case we evaluated the impact of the \(\sigma_R\) prior on the ABC estimates. xxx
2.7 Sensitivity of the selectivity estimate on the SRR
For this set of experiments, we evaluated the SRR curve given the past 20 years of selectivity estimates. From this we get a set of SRR curves that are conditioned on the selectivity estimate AND the MSY value. The script to run this set involved reading in the control file and modifying the option for which years to include for the selectivity estimate (can be specific years or a range of years from which to use).
Initial results showed that while the SRR curve was insensitive to the selectivity estimate (Figure 17), there could be large and variable impacts on the ABC estimates (Figure 18). This was due to the fact that the selectivity changes can shift to younger or older ages in some years. This also
2.8 Simulation testing the stock recruitment estimation
A simple simulation framework was set up to show how patterns of the Ricker SRR can be influenced by the available “points” used in the estimation. We start with the estimates of spawning biomass and recruits from the 2023 accepted model. As in the assessment, we selected the period from 1978-2022 and fit a Ricker SRR. We then replicated the estimation using random error about both the estimate of recruitment and spawning biomass. These “data” were then sampled with replacement. The
100 sets of data and resulting curves showed that the slope at the origin tends to be higher for these cases (are shown in (Figure 19). Note that these curves differ from the actual assessment since the fitting is done separately (we used the linear regression log recruits-per-spawning biomass vs spawning biomass). The point of this exercise is to show that extrapolating the fitted curves outside of the range of data (i.e., at smaller spawning stock sizes) can lead to very different and positively biased productivity estimates. The slope at the origin is a key parameter in the Ricker SRR and governs productivity estimates and consequently, \(F_{MSY}\) estimates. This suggests that applying the SRR estimates for management purposes (in Tier 1) may be inappropriate given this apparent potential for bias.
3 Incorporating natural mortality age arising from CEATTLE
In past assessments we have mentioned that a commonly adopted approach for stocks that are also included in multispecies trophic interaction models (reference ICES) it is considered best practice to include the estimates of natural mortality-at-age over time within the assessment model. We developed an option to include the 2023 CEATTLE estimates of natural mortality. This resulted slightly higher recruitment but lower spawning biomass in the near term (Figure 20). This is due to the higher natural mortality for most ages and years compared to the base 2023 model.
4 Pollock movement issues
Will include review of the publication and the plan for evaluating scenarios of alternative fishing mortality in the Russian zone. A research model evaluation may be presented in the 2024 assessment, depending on progress.
5 Ommitting early CPUE data and foreign fishery data
The SSC requested a model run where the early CPUE data were excluded. This was done and showed that the model was insensitive to the early CPUE data (Figure 22).
The SSC also noted “Catch-at-age data provided by foreign fishing agencies in the pre-Magnuson era were not produced using the same aging criteria as the AFSC age-and-growth program. Consideration should be given to removal of these data from the assessment. A sensitivity test should be done to evaluate the effects of data removal on the assessment.”
While these data are already downweighted by the effective sample size, we ran the model with the foreign catch-at-age data removed by setting the sample size to 1. Results showed that the model was sensitive to the removal of the foreign catch-at-age data for the early period but had little impact on near-term trends (Figure 23). Interestingly, the stock-recruit relationship was sensitive to the removal of the early period, presumably because of the data included in the fitting the the SRR between 1978 and 1991 being downweighted (Figure 24).
6 Tier 1 and ecosystem function evaluation
As noted, above, the SSC requested an evaluation of the ecosystem function as part of the SRR consideration and Tier 1 control rules within the FMP. Their comment was:
“The SSC would prefer not to make a risk table adjustment based on the difference from Tier 1 to Tier 3 again during the 2024 assessment cycle. The SSC requests that the next stock assessment bring back a new approach that may include development of a constant buffer based on factors extrinsic to the stock assessment (ecosystem function), or a better representation of the uncertainty in the Tier 1 and control rule calculations such that a reduction from maximum ABC is not needed every year.”
National Standard 1 (NS1) of the Magnuson-Stevens Act states that: “Conservation and management measures shall prevent overfishing while achieving, on a continuing basis, the optimum yield from each fishery for the United States fishing industry.” This standard involves balancing the competing policy objectives of preventing overfishing and achieving the optimum yield (OY). The specification of reference points such as maximum sustainable yield (MSY), OY, overfishing limit (OFL), acceptable biological catch (ABC), and annual catch limit (ACL) are central to U.S. fisheries management. The NS1 guidelines provide guidance on the specification of these reference points and the control rules used to establish limit and target catch levels. The NS1 guidelines require that each Fishery Management Council specify within their fishing management plans an ABC control rule that accounts for scientific uncertainty in the OFL and for the Council’s risk policy. The ABC cannot exceed the OFL. Beyond that, the guidelines provide flexibility in how ABC control rules can be specified. Many Councils have developed tiered ABC control rules. And many ABC control rules have risk policies that use the P* approach, where ABC is based on scientific uncertainty around the OFL and an acceptable probability of overfishing (P). The choice of P is often explicitly based on the status of the stock and other biological and ecological factors. Risk policies also include an element of policy choice between being risk adverse or risk tolerant, and implicit in this are social and economic considerations. This presentation will discuss the NS1 guidance on ABC control rules, highlight some of the flexibilities, and provide a few examples of how those flexibilities have been applied in practice.
The concern over the SSCs adjustment to the maximum permissible ABCs for EBS pollock stem from (in general) the magnitude of the ABCs and OFLs–they often exceed the 2 million t catch limit for the BSAI for all groundfish species combined. The over-arching TAC limit thus moderates the variability in advice from the Council to the Department of Commerce. As noted above, the SRR estimate is the main driver of the single-species ABCs. For context, the estimate for the long-term MSY is on the order of 2.2 million t of pollock catch. Over the past 4 decades, the actual EBS pollock catches have averaged about 1.3 million t.
As a point of curiosity, we considered inverting the SRR productivity estimate by posing the question “what SRR would give a long-term expected MSY of 1.3 million t.? Do our estimates of uncertainty in the SRR curve overlap substantially?
We thus added a feature of the model where one can provide a condition that the SRR be consistent with a specified MSY value.
As an experiment, we conditioned the SRR curve to have the MSY value set to 1.75 and 1.3 million t. We then compared those curves with the 2023 model specificaitons (Figure 25). When overplotted, the fit comparisons indicate somewhat worse fit to the available years of data (post 1977; Figure 26) but reasonable within the estimates of uncertainty. Here we conclude that the management advice (under Tier 1) is quite sensitive to relatively small apparent perturbations in the shape of the stock-recruitment curves.
6.0.1 Simulation testing an alternative management procedure
EBS pollock “maintain ecosystem function” catch-advice rule as requested by the SSC. One idea would be to evaluate the role of pollock as a key part of the forage base (say of 1-3 yr old pollock). A management goal with an explicit consideration of ecosystem function would be to avoid low levels. For example, if the forage base appeared to be close to say the lower 20th percentile from historical estimates, then a management procedure might include an adjustment that would occur then to avoid any further declines (Figure 27).
Practical aspects of such a management procedure might be prohibitive since in a projection scenario, information on those age groups would be limited. Consequently, we evaluated the prey base compared to spawning biomass (Figure 28). This figure shows that the relationship is poorly determined but that the forage component does tend to decline at lower spawning biomass levels. This suggests that by monitoring spawning biomass, there is potential linkages to downstream impacts of the main forage ages.
The fisheries management plan (FMP) for the BSAI is subject to the categorization of the stock assessment to obtain the maximum permissible ABC and OFL. For pollock, this falls on the determination of the appropriateness of the \(F_{MSY}\) and the probability distribution of that value (uncertainty). This relates directly to a number of key factors, including the selectivity, the SRR, and future weight-at-age (\(F_{MSY}\) applies to numbers of fish, but ABC is in biomass). within the FMP, an ABC can be set below the maximum permissible value. Without making an explicit amendment to the FMP, we propose an alternative semi-empirical approach. The first principle of this approach is to maintain the ecosystem function by contrasting from what’s been observed over the last several decades. Going forward, the catch advice could be adjusted based on the spawning biomass relative to the historical mean.
This would be a simple rule that would adjust the catch advice based on the spawning biomass relative to the historical mean. I.e., if the catch in the current year is say 1.2 million t, and the SSB next year is 30% above the mean, then with a regulator to dampen change, next year’s recommendation would be \(1.2 \times \sqrt{1.30} = 1.368\) million t. Similarly, if the SSB next year was only 75% of the mean value, the recommendation would be \(1.2 \times \sqrt{0.75} = 1.039\) million t. If the SSB stayed at 75% of the mean, then the following year would be \(1.039 \times \sqrt{0.75} = 08998\) million t.
In fact, the 2-million t overarching CAP on groundfish is often flagged as an having an ecosystem sustainability rationale. Bringing a similar rationale into focus for the pollock fishery such a management procedure would explicitly account for the ecosystem function and role of pollock. One way to evaluate such a procedure would be to use the available data on diet composition in a mass-balance ecosystem model. Here this could be set to compare models where pollock fishing is increased, decreased, and held near status quo.
Show the code
#|
M <- read_rep(here::here("2024", "runs", "test", "pm.rep"))
df <- data.frame(Year = 1964:2023, forage = rowSums(M$N), SSB = M$SSB[, 2]) |> mutate(rel_forage = forage / mean(forage), rel_SSB = SSB / mean(SSB))
qfor <- quantile(df$forage, c(0.05, .2))
df |> ggplot(aes(x = Year, y = forage)) +
geom_line() +
labs(title = "Total pollock forage (age 1-3)", x = "Year", y = "Abundance (thousands)") +
ggthemes::theme_few() +
xlim(c(1980, 2024)) +
geom_hline(yintercept = qfor[1], color = "red") +
geom_hline(yintercept = qfor[2], color = "blue", type = 2)
Show the code
df |>
pivot_longer(cols = 4:5, names_to = "var", values_to = "value") |>
filter(var == "rel_forage") |>
ggplot(aes(x = log(SSB), y = log(forage))) +
geom_point() +
geom_smooth() +
labs(title = "EBS Pollock ", x = "SSB", y = "Forage (ages 1 to 3 abundance)") +
ggthemes::theme_few() #+
# xlim(c(1980, 2024))
Show the code
df |> ggplot(aes(x = Year, y = SSB)) +
geom_line() +
labs(title = "Spawning biomass of pollock ", x = "Year", y = "Biomass (t) ") +
ggthemes::theme_few() +
xlim(c(1980, 2024)) +
geom_hline(yintercept = 2257, color = "red")
Show the code
#|
df1 <- NULL
for (resp in c(0.05, 0.5, 0.95)) {
for (i in 2000:2022) {
# get values from the time series
lastidx <- i - 1977 + 14
cat_mn <- mean(M$obs_catch[14:lastidx])
cat_this <- M$obs_catch[lastidx]
ssb_mn <- mean(M$SSB[14:lastidx, 2])
ssb_this <- (M$SSB[lastidx, 2])
ssb_next <- (M$SSB[lastidx + 1, 2])
# cat_next <- (ssb_next/ssb_this)^0.5 * cat_this
cat_next <- (ssb_next / ssb_mn)^resp * cat_this
df1 <- data.frame(Year = i, Mean_catch = cat_mn, Current_catch = cat_this, ABC = cat_next, ssb_mn = ssb_mn, ssb_this = ssb_this, ssb_next = ssb_next, resp = resp) |> rbind(df1)
}
}
df1 |>
pivot_longer(cols = c(3:5, 7), names_to = "var", values_to = "value") |>
mutate(Year = ifelse(var == "ABC", Year + 1, Year)) |>
ggplot(aes(x = Year, y = value, shape = var, color = var)) +
geom_point() +
geom_line() +
labs(title = "EBS Pollock ", x = "Year", y = "relative to mean") +
theme_bw() +
facet_grid(resp ~ .) +
xlim(c(2000, 2024))
Additionally, considering that humans are part of the ecosystem xxx
Such a management procedure would need to be tested using a full feedback loop system. Consequently, we updated the simulation feature of the assessment model to behave as an operating model. This required some simplifications on how the bottom-trawl survey covariance structure was applied, in addition to a couple of other issues (e.g., the random-effects derived estimates of weight-at-age).
6.0.2 Tests of future ecosystem states using EWE (Rpath) scenarios
To develop a more empirically based management procedure for pollock we recognize that this species is a keystone component for the Bering Sea. We thus posed the question on, given available understanding of ecosystem linkages, what might things look like if the fishery intensified or if it was removed altogether and compare that with “status quo” fishing.
The EBS pollock model has contained code to serve as an operating model for simulating datasets and be used as a self testing platform. Several modifications that have occurred since this was first implemented, including the addition of a new biomass index, required some code updates.
The model was run for 100 years under three scenarios: status
7 Bayesian diagnostics
Following the advice of Monnahan (2024) we performed multiple no-U-turn sampler (NUTS) chains and note that
that the potential scale reduction \(̂\hat{R}\) was <1.01 and the effective sample sizes were >400 for all parameters. For the sampling conducted there NUTS divergences were absent. As in past years, we used posterior predictive checks to validate models by confirming simulated data were consistent with the observations. Process error variances can be estimated jointly with random effects and other parameters when desired, and should be for important model components.
We also attempted the Pareto-smoothed importance sampling leave-one-out cross-validation (PSIS-LOO) method to assess model fit and found that the
More from Monnahan: [An approximate cross-validation technique called PSIS-LOO is the most practical tool for model selection, but can also provide important insights into model deficiencies.
I also recommended that model developers build and parameterize models to have minimal parameter correlations and marginal variances close to one, have options for diverse (multivariate) priors, do predictive modeling, and ensure that the tools comprising a workflow are accessible and straightforward for routine use. I review, adapt, and illustrate a Bayesian workflow on AD Model Builder and Stock Synthesis models, but these good practices apply to models from any software platform, including Template Model Builder and Stan. ]
8 Alternative software platforms
There is continued interest in using alternative software platforms for this assessment. A repository was developed for these alternatives here. The main reason for this is to provide options for upgrading the base software and providing some of the trade-offs between tailored assessments and general packages. In addition to the current model used for EBS pollock, alternaive platforms considered were:
Stock Synthesis 3: A very popular software platform
GOA pollock model: A customized program convertible between ADMB and TMB
SAM: A state-space model for age-structured assessments
AMAK: A general model assessment model developed to have flexible number of fisheries, indices etc.
WHAM: The Woods Hole Assessment Model (written in TMB…withdrawn from this presentation due to limits on time)
Each platform was intended to include as much of the configuration and baseline data from the pollock model as possible. Very little effort was made to do fine-scale bridging.
In subsequent sections we compare how the selectivity estimates compare, along with spawning biomass and recruitment.
8.0.1 Selectivity
Show the code
p<-Plot_Sel() + ggthemes::theme_few(base_size=9)
p
8.0.2 Selectivity by age
Show the code
p <- Plot_Sel_age()
ggplotly(p)8.0.3 SSB
Show the code
p<- Plot_SSB()
ggplotly(p)8.0.4 Stock recruitment relationship
Show the code
p<- Plot_SRR()
ggplotly(p)9 Additional AMAK runs
9.0.1 Run description
Runs with different selectivity assumptions where:
base: selectivity at age allowed to vary (sigma penalty=0.7)
cpue: As base but with the early CPUE data included
dbl_logistic: selectivity at age with TV selectivity parameters (3-parameter logistic)
| NLL Component | base | dbl_logistic | cpue |
|---|---|---|---|
| catch_like | 5.8 | 0.0 | 5.8 |
| age_like_fsh | 0.0 | 434.6 | 0.7 |
| length_like_fsh | 0.0 | 0.0 | 0.0 |
| sel_like_fsh | 0.0 | 342.9 | 1.0 |
| ind_like | 0.0 | 37.9 | 4.0 |
| age_like_ind | 0.0 | 81.3 | 0.2 |
| length_like_ind | 0.0 | 0.0 | 0.0 |
| sel_like_ind | 41.9 | 0.0 | 41.9 |
| rec_like | 0.0 | 11.1 | 4.8 |
| fpen | 0.0 | 0.0 | 0.0 |
| post_priors_indq | 0.0 | 0.2 | 0.0 |
| post_priors | 0.0 | 0.0 | 0.0 |
| residual | 0.0 | 0.0 | 0.0 |
| total | 0.0 | 860.4 | 10.7 |
9.0.2 Selectivity
9.0.3 Selectivity at age
9.0.4 SSB and recruitment
9.0.5 AMAK fit to indices
9.0.5.1 All
9.0.5.2 With CPUE
9.0.6 Stock recruitment relationship
Show the code
p<- Plot_SRR(df=am_ts)
ggplotly(p)10 Additional SS3 runs
10.0.1 Run description
Runs with different selectivity assumptions where:
- base: selectivity at age allowed to vary (sigma penalty=0.7)
- high: selectivity at age constrained (sigma penalty=0.05)
- mod: selectivity at age moderately constrained (sigma penalty=0.4)
- mix: selectivity at age moderately constrained for middle ages, high for older ages, loose for younger ages
10.0.2 Selectivity
10.0.3 Selectivity at age
10.0.4 SSB and recruitment
References
Monnahan, Cole C. 2024. “Toward Good Practices for Bayesian Data-Rich Fisheries Stock Assessments Using a Modern Statistical Workflow.” Fisheries Research 275: 107024. https://doi.org/https://doi.org/10.1016/j.fishres.2024.107024.